Moving Human Objects Recognition Research In Conplex Traiffc Environment

Multiple objects tracking and recognition technique with video image sequences incomplex traffic environment is one of the hot spots of intelligent traffic surveillance system inrecent years. It involves a wide range of technologies, including computer vision, imageprocessing and pattern recognition, artificial intelligence, communication etc. This papermainly focuses on multiple objects recognition, which is the important part of intelligentvideo surveillance system, and further research of moving human objects recognition inpassenger car video and related problems. After the comprehensive analysis of existing objectrecognition methods, this paper compares the traditional recognition methods and biomimeticpattern recognition(BPR) that starts with the way of knowing things like human beingsproposed by academician Wang Shoujue. Then, a novel moving human objects recognitionmethod of BPR based on high dimension space improved by fuzzy K-nearest neighborrecognition(FKNN) and another method bases on BPR and Adaboost are promoted, thesimulation experiment results verifies their efficiency and feasibility.Firstly, this paper studies the basic image processing techniques in complex trafficenvironment. Video image sequences preprocessing includes image de-noising and imageenhancement. Median filter and histogram equalization are chosen through the analysis andexperiments comparison of different methods. Then, it adopts the background subtractionwith improved adaptive Gaussian mixture model to deal with the dynamic scene changes incomplicated traffic environment and difficulties of moving objects extraction. In the stage ofmoving objects feature extraction, it presents common objects features in detail, especially themoving human body feature, and fused features of edge invariant moment, shape and gradientfeature of top view of human head contour image constitute the moving human objectscharacteristic values.To design human objects recognition algorithm, map the moving object feature into thehigh dimensional space according to BPR theory based on high dimensional space covering.Then analyze the distribution of moving human objects in the feature space to design triangleneuron for the moving objects covering implementation. A triangle neural network covering algorithm based on density selection is proposed to construct covering space. As theoverlapping space of different classes is easy to result in the error recognition, this paperemploys FKNN method to improve the further recognition. Experiment results indicate thatthe algorithm proposed this paper performs better than other methods, and it achieves highercorrect human objects recognition rate and reject rate, as well as reduce the false identification.On the other hand, it applies BPR into Adaboost algorithm for moving human objectsrecognition which has advantages both of Adaboost and BPR. Simulation results verify theeffectiveness of this method. That is to say, the two methods for moving human objects incomplex traffic environment both are effective and feasible.